The indexing queryset for review requests was hitting the same issue we

saw with datagrids and the API where a SELECT COUNT(*) would end up

performing a very expensive nested query, due to the usage of

DISTINCT. We have plans to remove DISTINCT after some further

testing, but probably not until Review Board 7.
However, we can now turn it off, if we don't mind the chance of

duplicates. That is fine for search indexing, since any duplicates will

just collapse into one entry.
This change turns off distinct queries when indexing, and also adds

another prefetch_related relation to pre-fetch diffsets. This reduces

the overall number of queries when indexing a batch of review requests.

Unit tests pass.
Tested building a full search index and inspecting all the generated

SQL queries, making sure we were performing simple queries and avoiding

excess queries.